app.py

import gradio as gr
import torch
import joblib
import numpy as np
from itertools import product
import torch.nn as nn
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
import io
from PIL import Image
from scipy.interpolate import interp1d

###############################################################################
# 1. MODEL DEFINITION
###############################################################################

class VirusClassifier(nn.Module):
    def __init__(self, input_shape: int):
        super(VirusClassifier, self).__init__()
        self.network = nn.Sequential(
            nn.Linear(input_shape, 64),
            nn.GELU(),
            nn.BatchNorm1d(64),
            nn.Dropout(0.3),
            nn.Linear(64, 32),
            nn.GELU(),
            nn.BatchNorm1d(32),
            nn.Dropout(0.3),
            nn.Linear(32, 32),
            nn.GELU(),
            nn.Linear(32, 2)
        )

    def forward(self, x):
        return self.network(x)

###############################################################################
# 2. FASTA PARSING & K-MER FEATURE ENGINEERING
###############################################################################

def parse_fasta(text):
    sequences = []
    current_header = None
    current_sequence = []
    for line in text.strip().split('\n'):
        line = line.strip()
        if not line: continue
        if line.startswith('>'):
            if current_header:
                sequences.append((current_header, ''.join(current_sequence)))
            current_header = line[1:]
            current_sequence = []
        else:
            current_sequence.append(line.upper())
    if current_header:
        sequences.append((current_header, ''.join(current_sequence)))
    return sequences

def sequence_to_kmer_vector(sequence: str, k: int = 4) -> np.ndarray:
    kmers = [''.join(p) for p in product("ACGT", repeat=k)]
    kmer_dict = {km: i for i, km in enumerate(kmers)}
    vec = np.zeros(len(kmers), dtype=np.float32)
    for i in range(len(sequence) - k + 1):
        kmer = sequence[i:i+k]
        if kmer in kmer_dict:
            vec[kmer_dict[kmer]] += 1
    total_kmers = len(sequence) - k + 1
    if total_kmers > 0:
        vec /= total_kmers
    return vec

###############################################################################
# 3. SHAP-VALUE (ABLATION) CALCULATION
###############################################################################

def calculate_shap_values(model, x_tensor):
    model.eval()
    with torch.no_grad():
        baseline_output = model(x_tensor)
        baseline_probs = torch.softmax(baseline_output, dim=1)
        baseline_prob = baseline_probs[0, 1].item()  # Prob of 'human'
        shap_values = []
        x_zeroed = x_tensor.clone()
        for i in range(x_tensor.shape[1]):
            original_val = x_zeroed[0, i].item()
            x_zeroed[0, i] = 0.0
            output = model(x_zeroed)
            probs = torch.softmax(output, dim=1)
            prob = probs[0, 1].item()
            shap_values.append(baseline_prob - prob)
            x_zeroed[0, i] = original_val
    return np.array(shap_values), baseline_prob

###############################################################################
# 4. PER-BASE SHAP AGGREGATION
###############################################################################

def compute_positionwise_scores(sequence, shap_values, k=4):
    kmers = [''.join(p) for p in product("ACGT", repeat=k)]
    kmer_dict = {km: i for i, km in enumerate(kmers)}
    seq_len = len(sequence)
    shap_sums = np.zeros(seq_len, dtype=np.float32)
    coverage = np.zeros(seq_len, dtype=np.float32)
    for i in range(seq_len - k + 1):
        kmer = sequence[i:i+k]
        if kmer in kmer_dict:
            val = shap_values[kmer_dict[kmer]]
            shap_sums[i:i+k] += val
            coverage[i:i+k] += 1
    with np.errstate(divide='ignore', invalid='ignore'):
        shap_means = np.where(coverage > 0, shap_sums / coverage, 0.0)
    return shap_means

###############################################################################
# 5. FIND EXTREME SHAP REGIONS
###############################################################################

def find_extreme_subregion(shap_means, window_size=500, mode="max"):
    n = len(shap_means)
    if n == 0: return (0, 0, 0.0)
    if window_size >= n:
        return (0, n, float(np.mean(shap_means)))
    csum = np.zeros(n + 1, dtype=np.float32)
    csum[1:] = np.cumsum(shap_means)
    best_start = 0
    best_sum = csum[window_size] - csum[0]
    best_avg = best_sum / window_size
    for start in range(1, n - window_size + 1):
        wsum = csum[start + window_size] - csum[start]
        wavg = wsum / window_size
        if mode == "max" and wavg > best_avg:
            best_avg = wavg; best_start = start
        elif mode == "min" and wavg < best_avg:
            best_avg = wavg; best_start = start
    return (best_start, best_start + window_size, float(best_avg))

###############################################################################
# 6. PLOTTING / UTILITIES
###############################################################################

def fig_to_image(fig):
    buf = io.BytesIO()
    fig.savefig(buf, format='png', bbox_inches='tight', dpi=150)
    buf.seek(0)
    img = Image.open(buf)
    plt.close(fig)
    return img

def get_zero_centered_cmap():
    colors = [(0.0, 'blue'), (0.5, 'white'), (1.0, 'red')]
    return mcolors.LinearSegmentedColormap.from_list("blue_white_red", colors)

def plot_linear_heatmap(shap_means, title="Per-base SHAP Heatmap", start=None, end=None):
    if start is not None and end is not None:
        local_shap = shap_means[start:end]
        subtitle = f" (positions {start}-{end})"
    else:
        local_shap = shap_means
        subtitle = ""
    if len(local_shap) == 0:
        local_shap = np.array([0.0])
    heatmap_data = local_shap.reshape(1, -1)
    min_val = np.min(local_shap)
    max_val = np.max(local_shap)
    extent = max(abs(min_val), abs(max_val))
    cmap = get_zero_centered_cmap()
    fig, ax = plt.subplots(figsize=(12, 1.8))
    cax = ax.imshow(heatmap_data, aspect='auto', cmap=cmap, vmin=-extent, vmax=extent)
    cbar = plt.colorbar(cax, orientation='horizontal', pad=0.25, aspect=40, shrink=0.8)
    cbar.ax.tick_params(labelsize=8)
    cbar.set_label('SHAP Contribution', fontsize=9, labelpad=5)
    ax.set_yticks([])
    ax.set_xlabel('Position in Sequence', fontsize=10)
    ax.set_title(f"{title}{subtitle}", pad=10)
    plt.subplots_adjust(bottom=0.25, left=0.05, right=0.95)
    return fig

def create_importance_bar_plot(shap_values, kmers, top_k=10):
    plt.rcParams.update({'font.size': 10})
    fig = plt.figure(figsize=(10, 5))
    indices = np.argsort(np.abs(shap_values))[-top_k:]
    values = shap_values[indices]
    features = [kmers[i] for i in indices]
    colors = ['#99ccff' if v < 0 else '#ff9999' for v in values]
    plt.barh(range(len(values)), values, color=colors)
    plt.yticks(range(len(values)), features)
    plt.xlabel('SHAP Value (impact on model output)')
    plt.title(f'Top {top_k} Most Influential k-mers')
    plt.gca().invert_yaxis()
    plt.tight_layout()
    return fig

def plot_shap_histogram(shap_array, title="SHAP Distribution in Region"):
    fig, ax = plt.subplots(figsize=(6, 4))
    ax.hist(shap_array, bins=30, color='gray', edgecolor='black')
    ax.axvline(0, color='red', linestyle='--', label='0.0')
    ax.set_xlabel("SHAP Value")
    ax.set_ylabel("Count")
    ax.set_title(title)
    ax.legend()
    plt.tight_layout()
    return fig

def compute_gc_content(sequence):
    if not sequence: return 0
    gc_count = sequence.count('G') + sequence.count('C')
    return (gc_count / len(sequence)) * 100.0

###############################################################################
# 7. MAIN ANALYSIS STEP (Gradio Step 1)
###############################################################################

def analyze_sequence(file_obj, top_kmers=10, fasta_text="", window_size=500):
    if fasta_text.strip():
        text = fasta_text.strip()
    elif file_obj is not None:
        try:
            with open(file_obj, 'r') as f:
                text = f.read()
        except Exception as e:
            return (f"Error reading file: {str(e)}", None, None, None, None)
    else:
        return ("Please provide a FASTA sequence.", None, None, None, None)

    sequences = parse_fasta(text)
    if not sequences:
        return ("No valid FASTA sequences found.", None, None, None, None)
    header, seq = sequences[0]

    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    try:
        state_dict = torch.load('model.pt', map_location=device, weights_only=True)
        model = VirusClassifier(256).to(device)
        model.load_state_dict(state_dict)
        scaler = joblib.load('scaler.pkl')
    except Exception as e:
        return (f"Error loading model/scaler: {str(e)}", None, None, None, None)

    freq_vector = sequence_to_kmer_vector(seq)
    scaled_vector = scaler.transform(freq_vector.reshape(1, -1))
    x_tensor = torch.FloatTensor(scaled_vector).to(device)

    shap_values, prob_human = calculate_shap_values(model, x_tensor)
    prob_nonhuman = 1.0 - prob_human
    classification = "Human" if prob_human > 0.5 else "Non-human"
    confidence = max(prob_human, prob_nonhuman)

    shap_means = compute_positionwise_scores(seq, shap_values, k=4)
    max_start, max_end, max_avg = find_extreme_subregion(shap_means, window_size, mode="max")
    min_start, min_end, min_avg = find_extreme_subregion(shap_means, window_size, mode="min")

    results_text = (
        f"Sequence: {header}\n"
        f"Length: {len(seq):,} bases\n"
        f"Classification: {classification}\n"
        f"Confidence: {confidence:.3f}\n"
        f"(Human Probability: {prob_human:.3f}, Non-human Probability: {prob_nonhuman:.3f})\n\n"
        f"---\n"
        f"**Most Human-Pushing {window_size}-bp Subregion**:\n"
        f"Start: {max_start}, End: {max_end}, Avg SHAP: {max_avg:.4f}\n\n"
        f"**Most Non-Human–Pushing {window_size}-bp Subregion**:\n"
        f"Start: {min_start}, End: {min_end}, Avg SHAP: {min_avg:.4f}"
    )

    kmers = [''.join(p) for p in product("ACGT", repeat=4)]
    bar_fig = create_importance_bar_plot(shap_values, kmers, top_kmers)
    bar_img = fig_to_image(bar_fig)

    heatmap_fig = plot_linear_heatmap(shap_means, title="Genome-wide SHAP")
    heatmap_img = fig_to_image(heatmap_fig)

    state_dict_out = {"seq": seq, "shap_means": shap_means}

    return (results_text, bar_img, heatmap_img, state_dict_out, header)

###############################################################################
# 8. SUBREGION ANALYSIS (Gradio Step 2)
###############################################################################

def analyze_subregion(state, header, region_start, region_end):
    if not state or "seq" not in state or "shap_means" not in state:
        return ("No sequence data found. Please run Step 1 first.", None, None)
    seq = state["seq"]
    shap_means = state["shap_means"]
    region_start = int(region_start)
    region_end = int(region_end)
    region_start = max(0, min(region_start, len(seq)))
    region_end = max(0, min(region_end, len(seq)))
    if region_end <= region_start:
        return ("Invalid region range. End must be > Start.", None, None)
    region_seq = seq[region_start:region_end]
    region_shap = shap_means[region_start:region_end]
    gc_percent = compute_gc_content(region_seq)
    avg_shap = float(np.mean(region_shap))
    positive_fraction = np.mean(region_shap > 0)
    negative_fraction = np.mean(region_shap < 0)
    if avg_shap > 0.05:
        region_classification = "Likely pushing toward human"
    elif avg_shap < -0.05:
        region_classification = "Likely pushing toward non-human"
    else:
        region_classification = "Near neutral (no strong push)"
    region_info = (
        f"Analyzing subregion of {header} from {region_start} to {region_end}\n"
        f"Region length: {len(region_seq)} bases\n"
        f"GC content: {gc_percent:.2f}%\n"
        f"Average SHAP in region: {avg_shap:.4f}\n"
        f"Fraction with SHAP > 0 (toward human): {positive_fraction:.2f}\n"
        f"Fraction with SHAP < 0 (toward non-human): {negative_fraction:.2f}\n"
        f"Subregion interpretation: {region_classification}\n"
    )
    heatmap_fig = plot_linear_heatmap(shap_means, title="Subregion SHAP", start=region_start, end=region_end)
    heatmap_img = fig_to_image(heatmap_fig)
    hist_fig = plot_shap_histogram(region_shap, title="SHAP Distribution in Subregion")
    hist_img = fig_to_image(hist_fig)
    return (region_info, heatmap_img, hist_img)

###############################################################################
# 9. COMPARISON ANALYSIS FUNCTIONS
###############################################################################

def normalize_shap_lengths(shap1, shap2, num_points=1000):
    """
    Normalize SHAP values to relative positions (0-1 scale).
    Each point represents a relative position in the sequence (e.g., 0.75 = 75% through sequence).
    """
    # Create relative position arrays (0 to 1)
    x1 = np.linspace(0, 1, len(shap1))
    x2 = np.linspace(0, 1, len(shap2))
    
    # Create normalized positions for comparison
    x_norm = np.linspace(0, 1, num_points)
    
    # Interpolate both sequences to the normalized positions
    shap1_interp = np.interp(x_norm, x1, shap1)
    shap2_interp = np.interp(x_norm, x2, shap2)
    
    return shap1_interp, shap2_interp

def compute_shap_difference(shap1_norm, shap2_norm):
    """Compute the SHAP difference between normalized sequences"""
    return shap2_norm - shap1_norm

def plot_comparative_heatmap(shap_diff, title="SHAP Difference Heatmap"):
    """
    Plot heatmap using relative positions (0-100%)
    """
    heatmap_data = shap_diff.reshape(1, -1)
    extent = max(abs(np.min(shap_diff)), abs(np.max(shap_diff)))
    
    fig, ax = plt.subplots(figsize=(12, 1.8))
    cmap = get_zero_centered_cmap()
    cax = ax.imshow(heatmap_data, aspect='auto', cmap=cmap, vmin=-extent, vmax=extent)
    
    # Create percentage-based x-axis ticks
    num_ticks = 5
    tick_positions = np.linspace(0, shap_diff.shape[0]-1, num_ticks)
    tick_labels = [f"{int(x*100)}%" for x in np.linspace(0, 1, num_ticks)]
    ax.set_xticks(tick_positions)
    ax.set_xticklabels(tick_labels)
    
    cbar = plt.colorbar(cax, orientation='horizontal', pad=0.25, aspect=40, shrink=0.8)
    cbar.ax.tick_params(labelsize=8)
    cbar.set_label('SHAP Difference (Seq2 - Seq1)', fontsize=9, labelpad=5)
    
    ax.set_yticks([])
    ax.set_xlabel('Relative Position in Sequence', fontsize=10)
    ax.set_title(title, pad=10)
    plt.subplots_adjust(bottom=0.25, left=0.05, right=0.95)
    
    return fig

def analyze_sequence_comparison(file1, file2, fasta1="", fasta2=""):
    """
    Compare two sequences using relative positions (0-1 scale)
    """
    # Analyze first sequence
    res1 = analyze_sequence(file1, top_kmers=10, fasta_text=fasta1, window_size=500)
    if isinstance(res1[0], str) and "Error" in res1[0]:
        return (f"Error in sequence 1: {res1[0]}", None, None)
    
    # Analyze second sequence
    res2 = analyze_sequence(file2, top_kmers=10, fasta_text=fasta2, window_size=500)
    if isinstance(res2[0], str) and "Error" in res2[0]:
        return (f"Error in sequence 2: {res2[0]}", None, None)

    shap1 = res1[3]["shap_means"]
    shap2 = res2[3]["shap_means"]
    
    # Normalize to relative positions
    shap1_norm, shap2_norm = normalize_shap_lengths(shap1, shap2)
    shap_diff = compute_shap_difference(shap1_norm, shap2_norm)
    
    # Calculate statistics
    avg_diff = np.mean(shap_diff)
    std_diff = np.std(shap_diff)
    max_diff = np.max(shap_diff)
    min_diff = np.min(shap_diff)
    
    threshold = 0.05
    substantial_diffs = np.abs(shap_diff) > threshold
    frac_different = np.mean(substantial_diffs)
    
    # Format output text
    len1_formatted = "{:,}".format(len(shap1))
    len2_formatted = "{:,}".format(len(shap2))
    classification1 = res1[0].split('Classification: ')[1].split('\n')[0].strip()
    classification2 = res2[0].split('Classification: ')[1].split('\n')[0].strip()
    
    comparison_text = (
        "Sequence Comparison Results:\n"
        f"Sequence 1: {res1[4]} (Length: {len1_formatted} bases)\n"
        f"Classification: {classification1}\n\n"
        f"Sequence 2: {res2[4]} (Length: {len2_formatted} bases)\n"
        f"Classification: {classification2}\n\n"
        "Comparison Statistics:\n"
        f"Average SHAP difference: {avg_diff:.4f}\n"
        f"Standard deviation: {std_diff:.4f}\n"
        f"Max difference: {max_diff:.4f} (Seq2 more human-like)\n"
        f"Min difference: {min_diff:.4f} (Seq1 more human-like)\n"
        f"Fraction of positions with substantial differences: {frac_different:.2%}\n\n"
        "Note: Comparisons shown at relative positions (0-100%) in each sequence\n"
        "Interpretation:\n"
        "- Red regions: Sequence 2 is more human-like at that relative position\n"
        "- Blue regions: Sequence 1 is more human-like at that relative position\n"
        "- White regions: Similar between sequences"
    )
    
    # Generate visualizations
    heatmap_fig = plot_comparative_heatmap(shap_diff)
    heatmap_img = fig_to_image(heatmap_fig)
    hist_fig = plot_shap_histogram(shap_diff, title="Distribution of SHAP Differences")
    hist_img = fig_to_image(hist_fig)
    
    return comparison_text, heatmap_img, hist_img

###############################################################################
# 10. BUILD GRADIO INTERFACE
###############################################################################

css = """
.gradio-container {
    font-family: 'IBM Plex Sans', sans-serif;
}
"""

with gr.Blocks(css=css) as iface:
    gr.Markdown("""
    # Virus Host Classifier
    **Step 1**: Predict overall viral sequence origin (human vs non-human) and identify extreme regions.  
    **Step 2**: Explore subregions to see local SHAP signals, distribution, GC content, etc.
    
    **Color Scale**: Negative SHAP = Blue, Zero = White, Positive = Red.
    """)
    
    with gr.Tab("1) Full-Sequence Analysis"):
        with gr.Row():
            with gr.Column(scale=1):
                file_input = gr.File(label="Upload FASTA file", file_types=[".fasta", ".fa", ".txt"], type="filepath")
                text_input = gr.Textbox(label="Or paste FASTA sequence", placeholder=">sequence_name\nACGTACGT...", lines=5)
                top_k = gr.Slider(minimum=5, maximum=30, value=10, step=1, label="Number of top k-mers to display")
                win_size = gr.Slider(minimum=100, maximum=5000, value=500, step=100, label="Window size for 'most pushing' subregions")
                analyze_btn = gr.Button("Analyze Sequence", variant="primary")
            with gr.Column(scale=2):
                results_box = gr.Textbox(label="Classification Results", lines=12, interactive=False)
                kmer_img = gr.Image(label="Top k-mer SHAP")
                genome_img = gr.Image(label="Genome-wide SHAP Heatmap (Blue=neg, White=0, Red=pos)")
        seq_state = gr.State()
        header_state = gr.State()
        analyze_btn.click(
            analyze_sequence,
            inputs=[file_input, top_k, text_input, win_size],
            outputs=[results_box, kmer_img, genome_img, seq_state, header_state]
        )

    with gr.Tab("2) Subregion Exploration"):
        gr.Markdown("""
        **Subregion Analysis**  
        Select start/end positions to view local SHAP signals, distribution, GC content, etc.
        The heatmap also uses the same Blue-White-Red scale.
        """)
        with gr.Row():
            region_start = gr.Number(label="Region Start", value=0)
            region_end = gr.Number(label="Region End", value=500)
            region_btn = gr.Button("Analyze Subregion")
        subregion_info = gr.Textbox(label="Subregion Analysis", lines=7, interactive=False)
        with gr.Row():
            subregion_img = gr.Image(label="Subregion SHAP Heatmap (B-W-R)")
            subregion_hist_img = gr.Image(label="SHAP Distribution (Histogram)")
        region_btn.click(
            analyze_subregion,
            inputs=[seq_state, header_state, region_start, region_end],
            outputs=[subregion_info, subregion_img, subregion_hist_img]
        )
    
    with gr.Tab("3) Comparative Analysis"):
        gr.Markdown("""
        **Compare Two Sequences**  
        Upload or paste two FASTA sequences to compare their SHAP patterns.
        The sequences will be normalized to the same length for comparison.
        
        **Color Scale**:  
        - Red: Sequence 2 is more human-like in this region  
        - Blue: Sequence 1 is more human-like in this region  
        - White: No substantial difference
        """)
        with gr.Row():
            with gr.Column(scale=1):
                file_input1 = gr.File(label="Upload first FASTA file", file_types=[".fasta", ".fa", ".txt"], type="filepath")
                text_input1 = gr.Textbox(label="Or paste first FASTA sequence", placeholder=">sequence1\nACGTACGT...", lines=5)
            with gr.Column(scale=1):
                file_input2 = gr.File(label="Upload second FASTA file", file_types=[".fasta", ".fa", ".txt"], type="filepath")
                text_input2 = gr.Textbox(label="Or paste second FASTA sequence", placeholder=">sequence2\nACGTACGT...", lines=5)
        compare_btn = gr.Button("Compare Sequences", variant="primary")
        comparison_text = gr.Textbox(label="Comparison Results", lines=12, interactive=False)
        with gr.Row():
            diff_heatmap = gr.Image(label="SHAP Difference Heatmap")
            diff_hist = gr.Image(label="Distribution of SHAP Differences")
        compare_btn.click(
            analyze_sequence_comparison,
            inputs=[file_input1, file_input2, text_input1, text_input2],
            outputs=[comparison_text, diff_heatmap, diff_hist]
        )
    
    gr.Markdown("""
    ### Interface Features
    - **Overall Classification** (human vs non-human) using k-mer frequencies.
    - **SHAP Analysis** to see which k-mers push classification toward or away from human.
    - **White-Centered SHAP Gradient**: 
      - Negative (blue), 0 (white), Positive (red), with symmetrical color range around 0. 
    - **Identify Subregions** with the strongest push for human or non-human.
    - **Subregion Exploration**: 
      - Local SHAP heatmap & histogram 
      - GC content 
      - Fraction of positions pushing human vs. non-human 
      - Simple logic-based classification
    - **Sequence Comparison**:
      - Compare two sequences to identify regions of difference
      - Normalized comparison to handle different sequence lengths
      - Statistical summary of differences
    """)

if __name__ == "__main__":
    iface.launch()